Update app.py

app.py

@@ -2,7 +2,7 @@
 Health Voice Assistant
 - Whisper ASR (incremental)
 - GPT health guidance (NON-diagnostic)
-- Auto Text-to-Speech (autoplay, no file saved)
+- Auto Text-to-Speech (autoplay, no button, no file saved)
 
 ⚠️ GENERAL guidance only – NOT medical advice
 """
@@ -14,11 +14,10 @@ import re
 import textwrap
 import base64
 from io import BytesIO
-from datetime import datetime
 
-import requests
 import numpy as np
 import torch
+import requests
 import streamlit as st
 
 from transformers import pipeline
@@ -26,13 +25,18 @@ from transformers.pipelines.audio_utils import ffmpeg_read
 from openai import OpenAI
 from gtts import gTTS
 
-# ================= PAGE CONFIG (MUST BE FIRST) =================
+# ================= PAGE CONFIG =================
 st.set_page_config(page_title="Health Voice Assistant", layout="wide")
 
 # ================= SESSION STATE =================
-for k in ["transcript", "gpt_text", "audio_html", "done"]:
-    if k not in st.session_state:
-        st.session_state[k] = None if k != "done" else False
+if "transcript" not in st.session_state:
+    st.session_state.transcript = None
+if "gpt_text" not in st.session_state:
+    st.session_state.gpt_text = None
+if "audio_html" not in st.session_state:
+    st.session_state.audio_html = None
+if "done" not in st.session_state:
+    st.session_state.done = False
 
 # ================= CONFIG =================
 MODEL_NAME = "openai/whisper-small.en"
@@ -51,10 +55,11 @@ device = 0 if torch.cuda.is_available() else -1
 st.title("🩺 Health Voice-Based Assistant")
 
 MEDICAL_DISCLAIMER = """
-⚠️ IMPORTANT MEDICAL DISCLAIMER  
-This assistant is NOT a medical professional.  
-It does NOT diagnose diseases or prescribe medication.  
-Information is for GENERAL guidance only.
+⚠️ **IMPORTANT MEDICAL DISCLAIMER**
+
+This assistant is **NOT** a medical professional.  
+It does **NOT** diagnose diseases or prescribe medication.  
+Information provided is for **GENERAL guidance only**.
 """
 
 # ================= SIDEBAR =================
@@ -62,12 +67,11 @@ st.sidebar.title("🎓 User Manual")
 st.sidebar.markdown(
     """
 1. Choose input method  
-2. Upload audio **or** record directly  
-3. Press **Transcribe**  
-4. App will:
-   - Analyze
-   - Give guidance
-   - Speak response
+2. Upload audio **or** record via web  
+3. App will automatically:
+   - Transcribe (or read transcript)
+   - Analyze with GPT
+   - Speak the guidance
 """
 )
 st.sidebar.markdown("---")
@@ -92,7 +96,6 @@ def load_pipeline():
         pass
     return pipe
 
-
 pipe = load_pipeline()
 
 # ================= GPT CLIENT =================
@@ -102,11 +105,15 @@ client = OpenAI(api_key=CHATGPT_API)
 def format_text(text: str, width: int = 90) -> str:
     text = re.sub(r"\s+", " ", text).strip()
     sentences = re.split(r"(?<=[.!?])\s+", text)
-    return "\n\n".join(
-        textwrap.fill(" ".join(sentences[i : i + 3]), width)
-        for i in range(0, len(sentences), 3)
-    )
-
+    paragraphs, buf = [], []
+    for s in sentences:
+        buf.append(s)
+        if len(buf) >= 3:
+            paragraphs.append(" ".join(buf))
+            buf = []
+    if buf:
+        paragraphs.append(" ".join(buf))
+    return "\n\n".join(textwrap.fill(p, width) for p in paragraphs)
 
 def merge_overlap(old: str, new: str, max_words: int = 10) -> str:
     if not old:
@@ -117,14 +124,35 @@ def merge_overlap(old: str, new: str, max_words: int = 10) -> str:
             return " ".join(a + b[k:])
     return old + " " + new
 
-
 def tts_bytes(text: str) -> BytesIO:
     mp3 = BytesIO()
     gTTS(text=text, lang="en").write_to_fp(mp3)
     mp3.seek(0)
     return mp3
 
+# ================= FIREBASE =================
+def fetch_firebase_transcript():
+    """
+    Read transcript text from Firebase Realtime Database
+    Path: /transcribe
+    """
+    try:
+        url = f"{FIREBASE_DB_URL}/transcribe.json"
+        res = requests.get(url, timeout=5)
+        res.raise_for_status()
+        data = res.json()
+
+        if not data:
+            return None
+
+        text = data.get("text", "").strip()
+        return text if text else None
 
+    except Exception as e:
+        st.error(f"Firebase read error: {e}")
+        return None
+
+# ================= GPT =================
 def run_gpt(transcript: str) -> str:
     prompt = f"""
 You are a NON-diagnostic health assistant.
@@ -150,34 +178,8 @@ TASK:
     )
     return res.choices[0].message.content.strip()
 
-
-# ================= FIREBASE =================
-def read_firebase_transcribe():
-    try:
-        url = f"{FIREBASE_DB_URL}/transcribe.json"
-        res = requests.get(url, timeout=5)
-        if res.status_code != 200:
-            return None
-
-        data = res.json()
-        if not data:
-            return None
-
-        text = data.get("text", "").strip()
-        if not text:
-            return None
-
-        return {
-            "text": text,
-            "length": data.get("length"),
-            "updated_at": data.get("updated_at"),
-        }
-    except Exception:
-        return None
-
-
 # ================= ASR =================
-def transcribe_audio(audio_bytes: bytes):
+def transcribe(audio_bytes: bytes):
     sr = pipe.feature_extractor.sampling_rate
     audio = ffmpeg_read(audio_bytes, sr)
 
@@ -190,10 +192,10 @@ def transcribe_audio(audio_bytes: bytes):
     chunk_samples = int(CHUNK_SEC * sr)
     overlap_samples = int(OVERLAP_SEC * sr)
 
+    total_chunks = max(1, (len(audio) + chunk_samples - 1) // chunk_samples)
     acc = ""
-    total = max(1, len(audio) // chunk_samples + 1)
 
-    for i in range(total):
+    for i in range(total_chunks):
         start = max(0, i * chunk_samples - overlap_samples)
         end = min(len(audio), (i + 1) * chunk_samples + overlap_samples)
 
@@ -203,106 +205,98 @@ def transcribe_audio(audio_bytes: bytes):
                 "forced_decoder_ids": getattr(pipe, "_forced_decoder_ids", None)
             },
         )
-        acc = merge_overlap(acc, result["text"].strip())
-        yield acc, i + 1, total
 
+        acc = merge_overlap(acc, result["text"].strip())
+        yield acc, i + 1, total_chunks
 
 # ================= INPUT MODE =================
-st.markdown("## 🎤 Audio Input Method")
+st.markdown("### 🎤 Input Mode")
 
-input_mode = st.radio(
+mode = st.radio(
     "Choose input method:",
-    ["Upload audio file", "Record audio (from Firebase)"],
-    horizontal=True,
+    ["Upload audio file", "Live recording (via Firebase)"],
 )
 
 # ================= MODE 2: FIREBASE =================
-if input_mode == "Record audio (from Firebase)":
-    st.info("📡 Checking Firebase for recorded audio...")
+if mode == "Live recording (via Firebase)":
+    st.info("Reading transcript from Firebase...")
+
+    firebase_text = fetch_firebase_transcript()
+
+    if firebase_text:
+        st.success("Transcript found. Processing with AI...")
+
+        st.session_state.audio_html = None
+        st.session_state.transcript = firebase_text
+
+        with st.spinner("Analyzing with GPT..."):
+            st.session_state.gpt_text = run_gpt(firebase_text)
 
-    fb = read_firebase_transcribe()
+        voice = tts_bytes(st.session_state.gpt_text)
+        b64 = base64.b64encode(voice.read()).decode()
+        st.session_state.audio_html = f"""
+        <audio autoplay>
+            <source src="data:audio/mp3;base64,{b64}" type="audio/mp3">
+        </audio>
+        """
 
-    # No data -> send user to the recording page (link + meta-refresh fallback)
-    if fb is None:
-        st.warning("No recorded data found in Firebase. Please record audio first.")
-        record_url = "https://nolist-ssp-recordoption.hf.space"
+        st.session_state.done = True
+
+    else:
+        st.warning("No transcript found. Redirecting to recorder...")
         st.markdown(
-            f"""
-            If you haven't recorded yet, open the recording page:
-            <a href="{record_url}" target="_blank" rel="noopener">Open recorder (new tab)</a>
-            <br><br>
-            The app will also attempt to redirect you automatically in 3 seconds.
-            <meta http-equiv="refresh" content="3; url={record_url}">
+            """
+            <meta http-equiv="refresh" content="0; url=https://nolist-ssp-recordoption.hf.space">
             """,
             unsafe_allow_html=True,
         )
-        st.stop()
-
-    # Data exists -> show transcript and a Continue button to run GPT+TTS
-    st.success("✅ Data received from Firebase")
-    st.markdown("### 📝 Transcript (from Firebase)")
-
-    st.markdown(f"**Length:** {fb.get('length', 'N/A')}")
-    if fb.get("updated_at"):
-        st.markdown(f"**Updated at:** {fb['updated_at']}")
-
-    st.markdown(f"```text\n{format_text(fb['text'])}\n```")
-
-    # Continue button: use same behavior as Upload->Transcribe, but with existing text
-    if st.button("Continue"):
-        try:
-            st.session_state.transcript = fb["text"]
-            st.session_state.gpt_text = run_gpt(fb["text"])
-
-            voice = tts_bytes(st.session_state.gpt_text)
-            b64 = base64.b64encode(voice.read()).decode()
-            st.session_state.audio_html = f"""
-            <audio autoplay>
-                <source src="data:audio/mp3;base64,{b64}" type="audio/mp3">
-            </audio>
-            """
-            st.success("Processed: GPT response generated and audio playing.")
-        except Exception as e:
-            st.error(f"Error processing Firebase transcript: {e}\n{traceback.format_exc()}")
-
-    st.stop()
 
 # ================= MODE 1: UPLOAD AUDIO =================
-uploaded = st.file_uploader(
-    "Upload audio (wav, mp3, m4a, ogg, flac)",
-    type=["wav", "mp3", "m4a", "ogg", "flac"],
-)
+if mode == "Upload audio file":
+    uploaded = st.file_uploader(
+        "Upload audio (wav, mp3, m4a, ogg, flac)",
+        type=["wav", "mp3", "m4a", "ogg", "flac"],
+    )
 
-if uploaded:
-    audio_bytes = uploaded.read()
-    st.audio(audio_bytes)
-
-    if st.button("Transcribe"):
-        try:
-            placeholder = st.empty()
-            progress = st.progress(0.0)
-
-            final_text = ""
-            for txt, i, total in transcribe_audio(audio_bytes):
-                final_text = txt
-                placeholder.markdown(f"```text\n{format_text(txt)}\n```")
-                progress.progress(i / total)
-
-            st.session_state.transcript = final_text
-            st.session_state.gpt_text = run_gpt(final_text)
-
-            voice = tts_bytes(st.session_state.gpt_text)
-            b64 = base64.b64encode(voice.read()).decode()
-            st.session_state.audio_html = f"""
-            <audio autoplay>
-                <source src="data:audio/mp3;base64,{b64}" type="audio/mp3">
-            </audio>
-            """
+    if uploaded:
+        audio_bytes = uploaded.read()
+        st.audio(audio_bytes)
+
+        if st.button("Transcribe"):
+            try:
+                st.session_state.audio_html = None
+
+                placeholder = st.empty()
+                progress = st.progress(0.0)
+
+                final_text = ""
+
+                for txt, i, total in transcribe(audio_bytes):
+                    final_text = txt
+                    placeholder.markdown(
+                        f"```text\n{format_text(txt)}\n```"
+                    )
+                    progress.progress(i / total)
+
+                st.session_state.transcript = final_text
+
+                with st.spinner("Analyzing with GPT..."):
+                    st.session_state.gpt_text = run_gpt(final_text)
+
+                voice = tts_bytes(st.session_state.gpt_text)
+                b64 = base64.b64encode(voice.read()).decode()
+                st.session_state.audio_html = f"""
+                <audio autoplay>
+                    <source src="data:audio/mp3;base64,{b64}" type="audio/mp3">
+                </audio>
+                """
+
+                st.session_state.done = True
 
-        except Exception as e:
-            st.error(f"{e}\n{traceback.format_exc()}")
+            except Exception as e:
+                st.error(f"Failed:\n{e}\n{traceback.format_exc()}")
 
-# ================= OUTPUT =================
+# ================= RENDER RESULT =================
 if st.session_state.transcript:
     st.markdown("### 📝 Transcript")
     st.markdown(f"```text\n{format_text(st.session_state.transcript)}\n```")
@@ -315,4 +309,4 @@ if st.session_state.audio_html:
     st.markdown(st.session_state.audio_html, unsafe_allow_html=True)
 
 st.markdown("---")
-st.markdown("⚠️ NOT medical advice. Seek professionals when needed.")
+st.markdown("⚠️ **NOT medical advice. Seek professional help when needed.**")